1271770 - Fallback to WARP if accelerated ANGLE fails

Status: RESOLVED FIXED

(Core :: Graphics: CanvasWebGL, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

Currently we don't do this. It seems like we should.

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

How's this related to bug 1260944?

Comment 2

[Jeff Muizelaar [:jrmuizel]]

Bug 1260944 made it possible to use WARP. FWIW, it looks like the machine I was seeing this on has an additional problem that makes WARP fail.

Comment 3

[Jeff Muizelaar [:jrmuizel]]

I think this bug is probably the reason for the remaining webgl (non performance caveat) failures on Windows 7

Comment 4

[Milan Sreckovic [:milan] (needinfo for best results)]

Benoit, something Ernest can cover?

Comment 5

[Kelsey Gilbert [:jgilbert]]

This is mostly solved by bug 1281050.

Comment 6

[Benoit Girard (:BenWa)]

Yes, I think this is a good bug.

Comment 7

[Benoit Girard (:BenWa)]

(In reply to Jeff Gilbert [:jgilbert] from comment #5)
> This is mostly solved by bug 1281050.

AIUI bug 1281050 would let us run with both ANGLE/Warp at the same time in the same process. I'm not sure that's desirable outside of testing environments. Is bug 1281050 really a blocker for this?

Comment 8

[Kelsey Gilbert [:jgilbert]]

(In reply to Benoit Girard (:BenWa) from comment #7)
> (In reply to Jeff Gilbert [:jgilbert] from comment #5)
> > This is mostly solved by bug 1281050.
> 
> AIUI bug 1281050 would let us run with both ANGLE/Warp at the same time in
> the same process. I'm not sure that's desirable outside of testing
> environments. Is bug 1281050 really a blocker for this?

Yes, because once we initialize EGL for non-WARP ANGLE, we can't re-initialize it for WARP ANGLE.

Comment 9

[Kelsey Gilbert [:jgilbert]]

If we aren't checking the ANGLE blocklist before initializing EGL, we can and should do that.
But once we initialize EGL, we're stuck with that ANGLE config. (including d3d9 vs d3d11)

Comment 10

[Ernest Yim [:ernest]]

Attachment: Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Review commit: https://reviewboard.mozilla.org/r/61570/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/61570/

Comment 11

[Ernest Yim [:ernest]]

Wrote a small patch for adding WARP fallback at initialization, will look into other locations where ANGLE may fail and see if we can fallback to WARP as discussed.

Comment 12

[Jeff Muizelaar [:jrmuizel]]

Early data from Nightly suggests that roughly 20% of our webgl failures would be fixed by this.

Comment 13

[Milan Sreckovic [:milan] (needinfo for best results)]

We're waiting on bug 1283594 before we want to land this, so that we can see the numbers change, or is there more that needs to be done to this patch?

Covering 20% of failures sounds great.

Comment 14

[Ernest Yim [:ernest]]

Currently the patch I wrote will only fallback at initialization, if we wait for bug 1281050 also, we can look into writing a patch to also fallback during runtime. But we can land the patch as is first, as the latter may take a bit more time and work.

Comment 15

[Benoit Girard (:BenWa)]

I don't think we need to block on bug 1283594. I believe we will still report the empty string so we should be able to study the data just fine.

Comment 16

[Benoit Girard (:BenWa)]

Actually the relationship needs to be the other way. Bug 1283594 is going to change heavily once this is landed because this will add ways we can fallback higher up in call graph.

Comment 17

[Ernest Yim [:ernest]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/61570/diff/1-2/

Comment 18

[Kelsey Gilbert [:jgilbert]]

Attachment: proposal

I think you want something like this instead.
We should try Accel unless we don't think it'll work.
After (maybe) trying Accel, we should fallback to WARP, except if we shouldn't be falling back to WARP.

Comment 19

[Ernest Yim [:ernest]]

With your suggestion, it will remove the overriding WARP preference (if ANGLE is successful)? Is this the intended behaviour?

Comment 20

[Ernest Yim [:ernest]]

Ignore the above comment, I lied.

Comment 21

[Benoit Girard (:BenWa)]

With this patch we might be trying WARP twice in some cases where it fails. Let's see if we can come up with better logic here.

Comment 22

[Ernest Yim [:ernest]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/61570/diff/2-3/

Comment 23

[Ernest Yim [:ernest]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/61570/diff/3-4/

Comment 24

[Jeff Muizelaar [:jrmuizel]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

https://reviewboard.mozilla.org/r/61570/#review61366

There's a machine in the office that would be good to test this on.

Comment 25

[Benoit Girard (:BenWa)]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

https://reviewboard.mozilla.org/r/61570/#review61702

Comment 26

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/af5a866739e9
Fallback to WARP if accelerated ANGLE fails r=BenWa,jrmuizel

Comment 27

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/af5a866739e9

Comment 28

[Jeff Muizelaar [:jrmuizel]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Approval Request Comment
[Feature/regressing bug #]: Not a regression
[User impact if declined]: WinVista+ users with GPUs that are old but haven't been explicitly blacklisted will not get WebGL. This looks to be the reason that Vista+ users don't get WebGL 71% of the time.
[Describe test coverage new/current, TreeHerder]: Has been on Nightly for a couple of days. However, it only exposes a path that we've already been testing on blacklisted GPUs.
[Risks and why]: I'd like to get as much testing as soon as possible, especially on the more diverse set of hardware that Aurora provides.

Comment 29

[Sylvestre Ledru [:Sylvestre]]

Comment on attachment 8766848 [details]
Bug 1271770 - Fallback to WARP if accelerated ANGLE fails

Improve WebGL support, taking it into aurora.

Comment 30

[Jeff Muizelaar [:jrmuizel]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/b28393478258

Comment 31

[Jeff Muizelaar [:jrmuizel]]

Release Note Request (optional, but appreciated)
[Why is this notable]: This increases our WebGL success rate noticeably (on beta the Win7 success rate went from 87% to 98%)
[Suggested wording]: Availability of WebGL increased to 98%+ on Windows 7 and newer.

Comment 32

[Ritu Kothari (:ritu) (Inactive, please n-i to RyanVM, jcristau, or pascal)]

Added to Fx50 release notes